Skip to main content

What the oracle is

The oracle is a read-only, publicly accessible projection layer over the b1e55ed event store. It answers one question:
Does this signal producer have verifiable history?
It is designed for AI agents and external systems to query before acting on a producer’s signals.

The provenance endpoint

  • No authentication required
  • Includes an anti-Goodhart header on every response:
Example (managed oracle):

MCP tool: b1e55ed_provenance_check

Agents can call the oracle through the MCP JSON-RPC tool b1e55ed_provenance_check. Via POST /api/v1/mcp:

Enabling it on your node

Oracle routes are enabled automatically when b1e55ed starts. No additional configuration is required. Your local oracle endpoint is available at:
No API key. No environment variables. If b1e55ed is running, the oracle is running.
The network-wide managed oracle is operated by Permanent Upper Class:
This is the oracle AI agents query by default. It aggregates provenance data across the network and handles contributor registration via the GitHub App. You do not need to run this yourself. Health check:

Self-hosted vs managed oracle

For most operators, the managed oracle is the right choice. Your node feeds data into it automatically when you contribute signals.

Advanced: running your own public oracle

This is only necessary if you are operating a fully sovereign, air-gapped deployment or building an independent oracle network. The GitHub App key is only held by the oracle server operator (Permanent Upper Class for the managed network). Operators who run b1e55ed do not need this key.
To run a public oracle with contributor registration:
  1. Create a GitHub App with issues: write permission on your repository
  2. Download the private key (.pem)
  3. Set the environment variable on your oracle server:
If running under systemd, use an environment file:
Then reference it in your service unit:
  1. Restart: